Diffusion Methods for Form Generalisation

In this section we focus our interest on techniques originating in computer vision. There we have the problem of distinguishing fine scale structures from coarse scale structures in digital images. A major step in solving this problem is related to the concept of a scale space. The classical example is the Gaussian scale space, which gives a linear method for the transition from fine to coarse scales. We demonstrate relations of this scale space to smoothing and averaging procedures. We show that the concept of a scale space is also applicable to form generalisations of digital elevation models (DEM). Anisotropy and inhomogeneity of DEM demand an application of nonlinear diffusion methods that have to be customised by steering parameters, which have to be interpretable in the context of DEM form generalisation. An example is given for elevation data from the Broeltal near Bonn.